Sensor Modeling , Calibration and Point Positioning with Terrestrial Panoramic Cameras

Several techniques have been used for terrestrial panoramic imaging. Known methods for panoramic imaging include: mosaicking/stitching of a rotated frame array CCD camera, mirror technology including single mirror and multi mirrors, near 180 degrees with large frame cameras or one shot with fish-eye lens and recently a linear array-based panoramic camera by horizontal rotation. Up to now, the technique of panorama production has mainly been used for pure imaging purposes, such as indoor imaging, landscape and cultural heritage recording, tourism advertising and image-based rendering and for efficient internet representations. Among the above panorama techniques, the linear array-based panoramic camera delivers a seamless high-resolution panoramic image with a Giga-pixel resolution in one shot. The camera consists of a linear array, which is mounted on a high precision turntable parallel to the rotation axis. The linear array sensor captures the scene by rotation of the turntable as a continuous set of vertical scan lines. The elegant image acquisition mode and the high information content of these panoramic images make them suitable candidates for quantitative image analysis. For accurate measurements a sophisticated camera calibration is an important prerequisite. However, due to intrinsic differences between the frame array pinhole camera model and a terrestrial linear array-based panoramic camera model, previously developed sensor models for frame array cameras cannot be used for the calibration of the terrestrial linear array-based panoramic cameras. Therefore a new sensor model is needed. We developed a sensor model for terrestrial linear array-based panoramic cameras by means of additional parameters, which models substantial deviations of a physical camera from the ideal one. The additional parameters model stationary and non-stationary systematic errors. The stationary systematic errors are related to: the lens, the configuration of the linear array with respect to the lens and the turntable, and the correction to the angular pixel size of the turntable. The source of non-stationary systematic errors is the dynamic mode of image acquisition. Systematic errors that are related to the dynamic behavior of the camera system are: non-equal angular pixel size and tumbling. The investigation of the modeling of these systematic errors is based on image space residual analysis and tumbling measurements by means of an inclinometer for the SpheroCam. Two terrestrial linear array-based panoramic cameras, the EYESCAN and the SpheroCam, are calibrated thorough self-calibration by the sensor model that was developed using additional parameters. Even though the system is highly dynamic a sub-pixel level of accuracy is obtained. The system’s accuracy for 3D point positioning is validated by use of specific testfields. We also investigate the minimum number of control points for the self-calibration. We extend the sensor model in order to calibrate terrestrial laser scanners providing laser intensity images, which operate similarly to panoramic cameras. Through the joint sensor model of terrestrial linear array-based panoramic cameras and terrestrial laser scanners, a laser scanner, Imager 5003, is calibrated. We achieve a sub-pixel level accuracy through self-calibration. A lot of control points are needed to determine the tumbling parameters through bundle adjustment. This makes the use of the sensor inconvenient in practice. Object space information, such as 3D straight-lines, provides extra conditions and reduces the number of control points for calibration and orientation. So far, straight-line information has been mainly used to determine interior orientation and additional parameters with single frame cameras. Due to the eccentricity of the projection center from the rotation axis, the acquired panoramic images do not have a single projection center. Therefore the mathematical model which has been used for single projection center cameras cannot be applied. We develop a new mathematical model for the processing of 3D straight-lines in panoramic images. We show that 3D straight-line information can be used in addition to tie points for calibration and orientation. This allows a full calibration and orientation without control points, which makes the use of the sensor more efficient. In addition to a sensor model, network design is considered to get the best possible accuracy. Network design considerations of panoramic cameras are assessed with respect to precision and reliability enhancement and the ability of the network for self-calibration. In terrestrial panoramic cameras the camera system is designed to have a leveled turntable, which reduces the mechanical errors of the camera system during rotation. However, this leads to restrictions in network design. Since the optical axis is always horizontal, the convergent concept cannot be realized in vertical direction. This loss in network flexibility must be compensated by other measures. We analyze several close-range networks of linear array-based panoramic camera stations through computer heuristic simulation in order to assess the precision and reliability. Joint networks of frame array CCD and linear array-based panoramic cameras are also compared with networks of linear array-based panoramic cameras only. We also investigate into the influence of different network configurations on the determination of additional parameters for self-calibration and point positioning. The accuracy and precision values of object points and the correlations of additional parameters with respect to the object point coordinates and the exterior orientation parameters are analyzed for this purpose. Networks of leveled and tilted linear array-based panoramic camera stations are analyzed by computer simulation. We show that with increasing tilt of camera stations the correlations of parameters decrease, especially the correlations of additional parameters with object space coordinates. We suggest tilted camera stations for self-calibration of linear array-based panoramic cameras and point positioning. We show the influence of datum definition on the solution vector (all unknown parameters) and the quality analysis matrices, which are computed from least squares bundle adjustment with an analytical proof. We also give numerical examples in addition to the analytical proof. The work that we present in this dissertation adds new and novel topics to the photogrammetric community. It provides preliminarily steps for further exploration such as measurement applications.